【2024年最新版】Windowsローカル環境でAIイラストを高速&大量生産 【Web UI Forge / Diffusers】【Stable Diffusion】

git --version

git clone https://github.com/lllyasviel/stable-diffusion-webui-forge.git

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge/repositories/stable-diffusion-webui-assets

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge/repositories/stable-diffusion-stability-ai

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge/repositories/generative-models

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge/repositories/k-diffusion

git config --global --add safe.directory D:/StableDiffusion/stable-diffusion-webui-forge/repositories/BLIP

cd /d D:\StableDiffusion\stable-diffusion-webui-forge
call webui-user.bat

python -m venv difenv

difenv\Scripts\activate.bat

deactivate

nvcc -V

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

pip install flax

pip install transformers

pip install diffusers["torch"] transformers

from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

pipe = StableDiffusionPipeline.from_pretrained('hakurei/waifu-diffusion')
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_attention_slicing()     
pipe=pipe.to("cuda")

#プロンプト
prompt = "masterpiece, best quality, super fine illustration, loli, an extremely cute and beautiful girl, detailed beautiful face, beautiful hair, beautiful eyes"

#ネガティブプロンプト
n_prompt="light, shadow, NSFW, worst quality, low quality, medium quality, deleted, lowres, comic,underboob,open_clothes, bad anatomy,strapless,bare_legs, bad hands, text,error,  missing fingers, extra digit, fewer digits, cropped , jpeg artifacts, signature, blurry, flat color, flat shading, nsfw, retro style, poor quality, bad face, bad fingers, bad anatomy, missing fingers, low res, cropped, signature, watermark, username, artist name, text"

# 処理の実行
image = pipe(
    prompt=prompt,
    negative_prompt=n_prompt,
    height=512,
    width=512,
    num_inference_steps=20, # =sample steps=何回処理を繰り返すか（多いほど精彩になる）
    guidance_scale=9 #＝CFG Scale＝どれほどプロンプトに従うか
).images[0]
    
image.save("save/test.png")

python generator-net.py

#generator.py

from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler

import datetime

model_path = "models/main/ambientmixAnAnime_v10.safetensors" #モデルが入ってるパス

pipe = StableDiffusionPipeline.from_single_file(model_path, load_safety_checker=False, extract_ema=True) 
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_attention_slicing()


prompt = "masterpiece, best quality, super fine illustration, loli, an extremely cute and beautiful girl, detailed beautiful face, beautiful hair, beautiful eyes"
each_prompt="navy maid clothes,maid_headdress,silver hair, upper body, solo" #さくやさん
n_prompt="light, shadow, NSFW, worst quality, low quality, medium quality, deleted, lowres, comic,underboob, open_clothes, bad anatomy,strapless,bare_legs, bad hands, text,error,  missing fingers, extra digit, fewer digits, cropped , jpeg artifacts, signature, blurry, flat color, flat shading, nsfw, retro style, poor quality, bad face, bad fingers, bad anatomy, missing fingers, low res, cropped, signature, watermark, username, artist name, text"

pipe=pipe.to("cuda")

for i in range(5): #rangeのカッコ内は生成枚数 
    image = pipe(
        prompt=prompt+","+each_prompt,
        negative_prompt=n_prompt,
        height=512,
        width=512,
        num_inference_steps=25,  #=sampling steps
        guidance_scale=9 #=CFG Scale
    ).images[0]

    image.save("save/"+datetime.datetime.now().strftime('%Y_%m_%d__%H_%M_%S')+".png") #日付・時刻をファイル名につける

#generator.py

from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
from diffusers.models import AutoencoderKL #VAEを使わなければこの行は不要

import datetime

model_path = "models/main/ambientmixAnAnime_v10.safetensors" #モデルのパス

#.safetensorsファイルのVAEを使うときは以下
#vae = AutoencoderKL.from_single_file("models/vae/diffusion_pytorch_model.safetensors") #VAEファイルのパス（VAEを使わなければこの行は不要）
#.binとconfig.jsonのVAEを使うときは以下
vae = AutoencoderKL.from_pretrained("models/vae/orangemixVAE") #VAEが入ってるフォルダのパス（VAEを使わなければこの行は不要）

pipe = StableDiffusionPipeline.from_single_file(model_path, load_safety_checker=False, extract_ema=True, vae=vae) #最後のvaeのところだけは、vaeを使わなければ不要。
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_attention_slicing()


prompt = "masterpiece, best quality, super fine illustration, loli, an extremely cute and beautiful girl, detailed beautiful face, beautiful hair, beautiful eyes"
each_prompt="navy maid clothes,maid_headdress,silver hair, upper body, solo" #さくやさん
n_prompt="light, shadow, NSFW, worst quality, low quality, medium quality, deleted, lowres, comic,underboob, open_clothes, bad anatomy,strapless,bare_legs, bad hands, text,error,  missing fingers, extra digit, fewer digits, cropped , jpeg artifacts, signature, blurry, flat color, flat shading, nsfw, retro style, poor quality, bad face, bad fingers, bad anatomy, missing fingers, low res, cropped, signature, watermark, username, artist name, text"

pipe=pipe.to("cuda")

for i in range(5): #rangeのカッコ内は生成枚数 
    image = pipe(
        prompt=prompt+","+each_prompt,
        negative_prompt=n_prompt,
        height=512,
        width=512,
        num_inference_steps=25,  #=sampling steps
        guidance_scale=9 #=CFG Scale
    ).images[0]

    image.save("save/"+datetime.datetime.now().strftime('%Y_%m_%d__%H_%M_%S')+".png") #日付・時刻をファイル名につける

python convert_pt_to_bin.py -h

python convert_pt_to_bin.py --vae_pt_path models\vae\orangemix.vae.pt --dump_path models\vae\orangemixVAE

# coding=utf-8
# Copyright 2023, Haofan Wang, Qixun Wang, All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

""" Conversion script for the LoRA's safetensors checkpoints. """

import argparse

import torch
from safetensors.torch import load_file

from diffusers import StableDiffusionPipeline

torch.set_default_tensor_type('torch.cuda.FloatTensor')
 
def load_safetensors_lora(pipeline, checkpoint_path, LORA_PREFIX_UNET="lora_unet", LORA_PREFIX_TEXT_ENCODER="lora_te", alpha=1):

    # load LoRA weight from .safetensors
    state_dict = load_file(checkpoint_path)

    visited = []

    # directly update weight in diffusers model
    for key in state_dict:
        # it is suggested to print out the key, it usually will be something like below
        # "lora_te_text_model_encoder_layers_0_self_attn_k_proj.lora_down.weight"

        # as we have set the alpha beforehand, so just skip
        if ".alpha" in key or key in visited:
            continue

        if "text" in key:
            layer_infos = key.split(".")[0].split(LORA_PREFIX_TEXT_ENCODER + "_")[-1].split("_")
            curr_layer = pipeline.text_encoder
        else:
            layer_infos = key.split(".")[0].split(LORA_PREFIX_UNET + "_")[-1].split("_")
            curr_layer = pipeline.unet

        # find the target layer
        temp_name = layer_infos.pop(0)
        while len(layer_infos) > -1:
            try:
                curr_layer = curr_layer.__getattr__(temp_name)
                if len(layer_infos) > 0:
                    temp_name = layer_infos.pop(0)
                elif len(layer_infos) == 0:
                    break
            except Exception:
                if len(temp_name) > 0:
                    temp_name += "_" + layer_infos.pop(0)
                else:
                    temp_name = layer_infos.pop(0)

        pair_keys = []
        if "lora_down" in key:
            pair_keys.append(key.replace("lora_down", "lora_up"))
            pair_keys.append(key)
        else:
            pair_keys.append(key)
            pair_keys.append(key.replace("lora_up", "lora_down"))

        # update weight
        if len(state_dict[pair_keys[0]].shape) == 4:
            weight_up = state_dict[pair_keys[0]].squeeze(3).squeeze(2).to(torch.float32)
            weight_down = state_dict[pair_keys[1]].squeeze(3).squeeze(2).to(torch.float32)
            curr_layer.weight.data += alpha * torch.mm(weight_up, weight_down).unsqueeze(2).unsqueeze(3)
        else:
            weight_up = state_dict[pair_keys[0]].to(torch.float32)
            weight_down = state_dict[pair_keys[1]].to(torch.float32)
            curr_layer.weight.data += alpha * torch.mm(weight_up, weight_down)            

        # update visited list
        for item in pair_keys:
            visited.append(item)

    return pipeline

#generator.py

from diffusers import StableDiffusionPipeline, DPMSolverMultistepScheduler
from diffusers.models import AutoencoderKL #VAEを使わなければこの行は不要
import convert_lora_safetensor_to_diffusers #LoRAを使わなければこの行は不要

import datetime

model_path = "models/main/ambientmixAnAnime_v10.safetensors" #モデルのパス
#.safetensorsファイルのVAEを使うときは以下
#vae = AutoencoderKL.from_single_file("models/vae/diffusion_pytorch_model.safetensors") #VAEファイルのパス（VAEを使わなければこの行は不要）
#.binとconfig.jsonのVAEを使うときは以下
vae = AutoencoderKL.from_pretrained("models/vae/orangemixVAE") #VAEが入ってるフォルダのパス（VAEを使わなければこの行は不要）
lora_path="models/lora/sakuya.safetensors" #LoRAが入ってるパス（LoRAを使わなければこの行は不要）

pipe = StableDiffusionPipeline.from_single_file(model_path, load_safety_checker=False, extract_ema=True, vae=vae) #最後のvaeのところだけは、vaeを使わなければ不要。
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe.enable_attention_slicing()

pipe = convert_lora_safetensor_to_diffusers.load_safetensors_lora(pipe, lora_path, alpha=0.7) #LoRAを使わなければこの行は不要

pipe=pipe.to("cuda")

prompt = "masterpiece, best quality, super fine illustration, loli, an extremely cute and beautiful girl, detailed beautiful face, beautiful hair, beautiful eyes"
each_prompt="navy maid clothes,maid_headdress,silver hair, upper body, solo" #さくやさん
n_prompt="light, shadow, NSFW, worst quality, low quality, medium quality, deleted, lowres, comic,underboob, open_clothes, bad anatomy,strapless,bare_legs, bad hands, text,error,  missing fingers, extra digit, fewer digits, cropped , jpeg artifacts, signature, blurry, flat color, flat shading, nsfw, retro style, poor quality, bad face, bad fingers, bad anatomy, missing fingers, low res, cropped, signature, watermark, username, artist name, text"

for i in range(5): #rangeのカッコ内は生成枚数 
    image = pipe(
        prompt=prompt+","+each_prompt,
        negative_prompt=n_prompt,
        height=512,
        width=512,
        num_inference_steps=25,  #=sampling steps
        guidance_scale=9 #=CFG Scale
    ).images[0]

    image.save("save/"+datetime.datetime.now().strftime('%Y_%m_%d__%H_%M_%S')+".png") #日付・時刻をファイル名につける

python generator.py